Chanler

Chanler

HomeArchives

"Dark Horse Redis Principles" Four, Memory Recovery

8
AI Translation
This post is translated from Chinese into English through AI.View Original
AI-generated summary
Redis relies on in-memory storage for high performance, but the memory size of a single node should not be too large to avoid impacting persistence and replication. The maximum memory can be configured (e.g., `maxmemory 1gb`), and once this limit is reached, no more data can be stored, leading to the need for memory reclamation strategies. ### Expiration Policies - **TTL (Time to Live)**: Keys can have an expiration time set, and accessing an expired key will return nil. - **Lazy Deletion**: Keys are not immediately deleted upon expiration; instead, they are checked for expiration when accessed. - **Periodic Deletion**: A scheduled task periodically samples and deletes expired keys. This can happen during Redis initialization or before each event loop. ### Eviction Policies When Redis memory usage hits a set threshold, it actively selects keys to delete to free up memory. If memory is insufficient during command processing and `maxmemory` is set (and not executing a Lua script), Redis attempts to evict memory. Eviction strategies include: - **LRU (Least Recently Used)**: Keys that haven't been accessed for the longest time are deleted. - **LFU (Least Frequently Used)**: Keys that are accessed the least are deleted. The choice of which keys to delete depends on their TTL, access frequency, and usage patterns.

Memory Reclamation#

The main reason for Redis's strong performance is its memory-based storage. However, the memory size of a single-node Redis should not be too large, as it can affect persistence or master-slave synchronization performance.

In the configuration file, you can set the maximum memory for Redis using maxmemory <bytes>, such as maxmemory 1gb. When the memory reaches the limit, it can no longer store data.

This leads to the memory reclamation strategy.

Expiration Strategy#

expire <key> <expireTime> can set a time-to-live (TTL) for a key. When it expires, accessing it will return nil.

image.png|500

image.png|500

How does Redis know if a key has expired?
It uses two dictionaries to record key-value and key-ttl separately.

Does it delete immediately when the TTL expires?
For example:
Lazy deletion
Periodic deletion

Lazy Deletion#

It does not delete immediately when the TTL expires, but checks the key's lifespan when accessing the key. If it is expired, it deletes it.

image.png|500

Periodic Deletion#

Periodic deletion: This is done through a scheduled task that periodically samples some expired keys and then executes deletion. There are two execution cycles:

  • Redis sets a scheduled task serverCron() during initialization, which cleans expired keys at the frequency of server.hz, in SLOW mode.
  • Before each event loop in Redis, the beforeSleep() function is called to clean expired keys, in FAST mode.

image.png|500

image.png|500

Eviction Strategy#

Memory eviction: When Redis memory usage reaches the set threshold, it actively selects some keys to delete in order to free up more memory.

When processing commands, if memory is insufficient, and maxmemory is set and it is not a Lua script being executed, it will first attempt to evict memory.

image.png|500

Eviction Strategy

image.png|500

How to know LRU and LFU?

image.png|500

Eviction Strategy Flowchart

For TTL, the longer the lifespan, the more it should be deleted; for LRU, the less recently accessed, the more it should be deleted; for LFU, the less frequently accessed, the more it should be deleted.

image.png|500

This article is synchronized and updated to xLog by Mix Space. The original link is https://blog.0xling.cyou/posts/redis/redis-6

Loading...
Ownership of this post data is guaranteed by blockchain and smart contracts to the creator alone.